Welcome to RealityFuse XL!
Hello everyone, this is my first SDXL model. With this model, it is possible to create more realistic photographs. The aim of the model is to create realistic images and to produce as high quality photographs as possible. This model is still under development and is being improved every day. If you want to create realistic photos, you can try my model and send me feedback.
My plans for the near future:
The model is now working very well at +15 step. But I plan to reduce it to 5 or 10 in the near future, so you can generate faster. I will do my best to run faster and more stable. I currently have 3060ti and ryzen 5 3600x. I generate 1 photo in about 30 seconds. When upscaled, this time increases twice. Upscaled photos look better quality and clearer. So I recommend using a nice upscale model.
⭐ Don't forget to support me if you like my models! ⭐
⭐ https://www.buymeacoffee.com/cemrek ⭐
Recommended settings and creation settings:
Upscaler: 4x ULTRA SHARP
Denoising strength: 0.25 ~ 0.35
Hires steps: 10 ~
Upscale by: 1.5 ~
Sampling Steps: 15 ~ (Higher values best for realism)
Sampling Method: DPM++ SDE Karras
CFG Scale: 2 ~ 3
Height x Weight: 1152x896 or higher
Negative Words: (canvas,white border,black border,frame), bad-hands-5, ((watermark,3d,cartoon,paint,draw,illustration,anime,render,3d render)),(muted colors:1.4),pixelated, low res, jpeg artifacts, Compression artifacts, bad art, worst quality, low resolution, low quality, plastic, fake, bad limbs, conjoined, featureless, bad features, incorrect objects, watermark, signature, logo
Positive Words: 8k wallpaper,ultra detailed,close up, film effect, grain,detailed,hdr,cinemtic shoot,intricated skin,realistic,photorealistic, rim lights, dark shadows, cinematic scene,depth of field,epic
Known bugs:
-Sometimes faces are not clear in distant photographs. I recommend using ADetailer for this.
-Sometimes the same photos always generate. This will be fixed soon.
-Sometimes the faces are not symmetrical. Please generate again to get rid of this.
This model is under development. please remember to save the model to your favourites so that you will be notified as the model develops.
For business inquires, commercial licensing, custom models, and consultation contact me under [email protected]
Description
v1 WIP
FAQ
Comments (10)
Is this trained from the SDXL base model, or did you use one of the custom photorealistic models as a starting point?
Hey, I adapted the three sd 1.5 models I used before to the xl model. Then I combined it with the xl base model. and trained.
@CemreK Adapting SD 1.5 models to SDXL? I didn't even know that's possible. Do you have a link to a howto?
@steamrick X-Adapter search on github :) if you have trouble add me on discord : priwwe
@CemreK So if I understand this correctly, you're taking a Lora extracted from an SD1.5 base model and fusing it with an SDXL base model using X-Adapter?
op, describe more please ?
this is just another juggernaut-merge
No, not from the description in the previous comment. Maybe look around a bit before spouting opinions. It is not fair to the creator.
They mentioned in the other comment that this is a merge of 1.5 models into an sdxl using x-adapter.
Doesn't sound like a juggernaut merge at all to me. But compliments could be taken if that the quality produced makes you assume such.
i have some juggernaut-merges and one of them produces not just the same images for a seed like this one. it produces identical images... no other model does that. i think that model was a merge with fluenty and swiper (train differences).
you should not just read the description, you should try it out before you have an opinion
@adrianamotnik953 If that's true then it really shouldn't be labled as trained @CemreK